xref: /llvm-project/llvm/test/CodeGen/AMDGPU/v_cmp_gfx11.ll (revision 6548b6354d1d990e1c98736f5e7c3de876bedc8e)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=amdgcn-amd-amdhsa -mattr=+wavefrontsize64 --global-isel=0 -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=CHECK %s
3
4define amdgpu_kernel void @icmp_test() {
5; CHECK-LABEL: icmp_test:
6; CHECK:       ; %bb.0: ; %entry
7; CHECK-NEXT:    v_cmp_eq_u16_e64 s[0:1], 0, 0
8; CHECK-NEXT:    v_mov_b32_e32 v1, 0
9; CHECK-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
10; CHECK-NEXT:    s_cmp_eq_u64 s[0:1], 0
11; CHECK-NEXT:    s_cselect_b64 s[0:1], -1, 0
12; CHECK-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s[0:1]
13; CHECK-NEXT:    ds_store_b32 v1, v0
14; CHECK-NEXT:    s_endpgm
15entry:
16  %icmp.intr = tail call i64 @llvm.amdgcn.icmp.i64.i16(i16 0, i16 0, i32 32)
17  %cmp0 = icmp eq i64 %icmp.intr, 0
18  %zext0 = zext i1 %cmp0 to i32
19  store i32 %zext0, ptr addrspace(3) null, align 2147483648
20  ret void
21}
22
23define amdgpu_kernel void @fcmp_test(half %x, half %y) {
24; CHECK-LABEL: fcmp_test:
25; CHECK:       ; %bb.0: ; %entry
26; CHECK-NEXT:    s_load_b32 s0, s[4:5], 0x0
27; CHECK-NEXT:    v_mov_b32_e32 v1, 0
28; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
29; CHECK-NEXT:    s_lshr_b32 s1, s0, 16
30; CHECK-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
31; CHECK-NEXT:    v_cmp_le_f16_e64 s[0:1], s0, s1
32; CHECK-NEXT:    s_cmp_eq_u64 s[0:1], 0
33; CHECK-NEXT:    s_cselect_b64 s[0:1], -1, 0
34; CHECK-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
35; CHECK-NEXT:    v_cndmask_b32_e64 v0, 0, 1, s[0:1]
36; CHECK-NEXT:    ds_store_b32 v1, v0
37; CHECK-NEXT:    s_endpgm
38entry:
39  %fcmp.intr = tail call i64 @llvm.amdgcn.fcmp.i64.f16(half %x, half %y, i32 5)
40  %cmp0 = icmp eq i64 %fcmp.intr, 0
41  %zext0 = zext i1 %cmp0 to i32
42  store i32 %zext0, ptr addrspace(3) null, align 2147483648
43  ret void
44}
45
46define amdgpu_kernel void @ballot_test(half %x, half %y) {
47; CHECK-LABEL: ballot_test:
48; CHECK:       ; %bb.0:
49; CHECK-NEXT:    s_load_b32 s0, s[4:5], 0x0
50; CHECK-NEXT:    v_mov_b32_e32 v2, 0
51; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
52; CHECK-NEXT:    s_lshr_b32 s1, s0, 16
53; CHECK-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_1)
54; CHECK-NEXT:    v_cmp_eq_f16_e64 s[0:1], s0, s1
55; CHECK-NEXT:    v_mov_b32_e32 v0, s0
56; CHECK-NEXT:    s_delay_alu instid0(VALU_DEP_2)
57; CHECK-NEXT:    v_mov_b32_e32 v1, s1
58; CHECK-NEXT:    ds_store_b64 v2, v[0:1]
59; CHECK-NEXT:    s_endpgm
60  %cmp = fcmp oeq half %x, %y
61  %ballot = tail call i64 @llvm.amdgcn.ballot.i64(i1 %cmp)
62  store i64 %ballot, ptr addrspace(3) null, align 2147483648
63  ret void
64}
65
66declare i64 @llvm.amdgcn.icmp.i64.i16(i16, i16, i32 immarg) #0
67
68declare i64 @llvm.amdgcn.fcmp.i64.f16(half, half, i32 immarg) #0
69
70declare i64 @llvm.amdgcn.ballot.i64(i1) #0
71
72attributes #0 = { convergent nounwind readnone willreturn }
73